Model – Data synthesis using Bayesian analysis

To the modeller, there are a wide range of potential valid models to choose from to explain any particular dataset. All these models have multiple parameters to be fitted. How to calibrate these parameters? A method championed by Vrugt is running multiple search algorithms concurrently that learn from each other through information exchange using a common population of points. This is an elegant way of exploiting the strengths of individual search methods to address the No Free Lunch Theorem. Numerical experiments demonstrate that AMALGAM and DREAM outperform existing evolutionary and MCMC approaches. These methods are especially designed to take maximum advantage of distributed computing facilities, and are currently being applied to a wide range of inference problems. Vrugt will talk about some of these methods, and show a bunch of applications from biology, ecology, hydrology, geophysics, aerosol cloud interactions, etc.. to illustrate their usefulness.

About the speaker:

Vrugt was born in Amsterdam, The Netherlands. He graduated from The University of Amsterdam where he received his MS (Cum Laude) and PhD degree (Cum Laude). During his PhD, Vrugt developed a wide variety of new concepts and algorithms for analysis of the information content of experimental data and Bayesian, and multi-criteria treatment and quantification of model parameter, state, and predictive uncertainty. Upon completion of his doctorate, Vrugt started working as a Director’s Postdoctoral Fellow at the Los Alamos National Laboratory (LANL). Here he developed a novel concept of self-adaptve multimethod evolutionary optimization for single and multiple objective parameter estimation, and an adaptive Markov Chain Monte Carlo (MCMC) scheme (DREAM) for sampling of complex high-dimensional posterior distributions. In 2010, Vrugt was appointed Assistant Professor of Civil and Environmental Engineering at the University of California, Irvine.

My research group uses optimality principles, Bayesian statistics, Monte Carlo simulation and evolutionary strategies to better analyze the mismatch between models and data and help improve theory, understanding and predictability of environmental systems. We regularly develop new methods and use parallel computing to solve the most complex and computationally demanding (inverse) problems. We draw inspiration from emerging model-data synthesis problems in surface hydrology, soil physics, hydrogeophysics, hydrometeorology, and geophysics. Most recently, we have started working on ecological inference problems and use optimality hypotheses to understand why plants grow in certain patterns, and why birds choose certain migratory flight routes.